69c9bab89329989d874523a8591137e93d549f3a
[libav.git] / libavresample / resample.c
1 /*
2 * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
4 *
5 * This file is part of Libav.
6 *
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "libavutil/common.h"
23 #include "libavutil/libm.h"
24 #include "libavutil/log.h"
25 #include "internal.h"
26 #include "resample.h"
27 #include "audio_data.h"
28
29 struct ResampleContext {
30 AVAudioResampleContext *avr;
31 AudioData *buffer;
32 uint8_t *filter_bank;
33 int filter_length;
34 int ideal_dst_incr;
35 int dst_incr;
36 int index;
37 int frac;
38 int src_incr;
39 int compensation_distance;
40 int phase_shift;
41 int phase_mask;
42 int linear;
43 enum AVResampleFilterType filter_type;
44 int kaiser_beta;
45 double factor;
46 void (*set_filter)(void *filter, double *tab, int phase, int tap_count);
47 void (*resample_one)(struct ResampleContext *c, int no_filter, void *dst0,
48 int dst_index, const void *src0, int src_size,
49 int index, int frac);
50 };
51
52
53 /* double template */
54 #define CONFIG_RESAMPLE_DBL
55 #include "resample_template.c"
56 #undef CONFIG_RESAMPLE_DBL
57
58 /* float template */
59 #define CONFIG_RESAMPLE_FLT
60 #include "resample_template.c"
61 #undef CONFIG_RESAMPLE_FLT
62
63 /* s32 template */
64 #define CONFIG_RESAMPLE_S32
65 #include "resample_template.c"
66 #undef CONFIG_RESAMPLE_S32
67
68 /* s16 template */
69 #include "resample_template.c"
70
71
72 /* 0th order modified bessel function of the first kind. */
73 static double bessel(double x)
74 {
75 double v = 1;
76 double lastv = 0;
77 double t = 1;
78 int i;
79
80 x = x * x / 4;
81 for (i = 1; v != lastv; i++) {
82 lastv = v;
83 t *= x / (i * i);
84 v += t;
85 }
86 return v;
87 }
88
89 /* Build a polyphase filterbank. */
90 static int build_filter(ResampleContext *c)
91 {
92 int ph, i;
93 double x, y, w, factor;
94 double *tab;
95 int tap_count = c->filter_length;
96 int phase_count = 1 << c->phase_shift;
97 const int center = (tap_count - 1) / 2;
98
99 tab = av_malloc(tap_count * sizeof(*tab));
100 if (!tab)
101 return AVERROR(ENOMEM);
102
103 /* if upsampling, only need to interpolate, no filter */
104 factor = FFMIN(c->factor, 1.0);
105
106 for (ph = 0; ph < phase_count; ph++) {
107 double norm = 0;
108 for (i = 0; i < tap_count; i++) {
109 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
110 if (x == 0) y = 1.0;
111 else y = sin(x) / x;
112 switch (c->filter_type) {
113 case AV_RESAMPLE_FILTER_TYPE_CUBIC: {
114 const float d = -0.5; //first order derivative = -0.5
115 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
116 if (x < 1.0) y = 1 - 3 * x*x + 2 * x*x*x + d * ( -x*x + x*x*x);
117 else y = d * (-4 + 8 * x - 5 * x*x + x*x*x);
118 break;
119 }
120 case AV_RESAMPLE_FILTER_TYPE_BLACKMAN_NUTTALL:
121 w = 2.0 * x / (factor * tap_count) + M_PI;
122 y *= 0.3635819 - 0.4891775 * cos( w) +
123 0.1365995 * cos(2 * w) -
124 0.0106411 * cos(3 * w);
125 break;
126 case AV_RESAMPLE_FILTER_TYPE_KAISER:
127 w = 2.0 * x / (factor * tap_count * M_PI);
128 y *= bessel(c->kaiser_beta * sqrt(FFMAX(1 - w * w, 0)));
129 break;
130 }
131
132 tab[i] = y;
133 norm += y;
134 }
135 /* normalize so that an uniform color remains the same */
136 for (i = 0; i < tap_count; i++)
137 tab[i] = tab[i] / norm;
138
139 c->set_filter(c->filter_bank, tab, ph, tap_count);
140 }
141
142 av_free(tab);
143 return 0;
144 }
145
146 ResampleContext *ff_audio_resample_init(AVAudioResampleContext *avr)
147 {
148 ResampleContext *c;
149 int out_rate = avr->out_sample_rate;
150 int in_rate = avr->in_sample_rate;
151 double factor = FFMIN(out_rate * avr->cutoff / in_rate, 1.0);
152 int phase_count = 1 << avr->phase_shift;
153 int felem_size;
154
155 if (avr->internal_sample_fmt != AV_SAMPLE_FMT_S16P &&
156 avr->internal_sample_fmt != AV_SAMPLE_FMT_S32P &&
157 avr->internal_sample_fmt != AV_SAMPLE_FMT_FLTP &&
158 avr->internal_sample_fmt != AV_SAMPLE_FMT_DBLP) {
159 av_log(avr, AV_LOG_ERROR, "Unsupported internal format for "
160 "resampling: %s\n",
161 av_get_sample_fmt_name(avr->internal_sample_fmt));
162 return NULL;
163 }
164 c = av_mallocz(sizeof(*c));
165 if (!c)
166 return NULL;
167
168 c->avr = avr;
169 c->phase_shift = avr->phase_shift;
170 c->phase_mask = phase_count - 1;
171 c->linear = avr->linear_interp;
172 c->factor = factor;
173 c->filter_length = FFMAX((int)ceil(avr->filter_size / factor), 1);
174 c->filter_type = avr->filter_type;
175 c->kaiser_beta = avr->kaiser_beta;
176
177 switch (avr->internal_sample_fmt) {
178 case AV_SAMPLE_FMT_DBLP:
179 c->resample_one = resample_one_dbl;
180 c->set_filter = set_filter_dbl;
181 break;
182 case AV_SAMPLE_FMT_FLTP:
183 c->resample_one = resample_one_flt;
184 c->set_filter = set_filter_flt;
185 break;
186 case AV_SAMPLE_FMT_S32P:
187 c->resample_one = resample_one_s32;
188 c->set_filter = set_filter_s32;
189 break;
190 case AV_SAMPLE_FMT_S16P:
191 c->resample_one = resample_one_s16;
192 c->set_filter = set_filter_s16;
193 break;
194 }
195
196 felem_size = av_get_bytes_per_sample(avr->internal_sample_fmt);
197 c->filter_bank = av_mallocz(c->filter_length * (phase_count + 1) * felem_size);
198 if (!c->filter_bank)
199 goto error;
200
201 if (build_filter(c) < 0)
202 goto error;
203
204 memcpy(&c->filter_bank[(c->filter_length * phase_count + 1) * felem_size],
205 c->filter_bank, (c->filter_length - 1) * felem_size);
206 memcpy(&c->filter_bank[c->filter_length * phase_count * felem_size],
207 &c->filter_bank[(c->filter_length - 1) * felem_size], felem_size);
208
209 c->compensation_distance = 0;
210 if (!av_reduce(&c->src_incr, &c->dst_incr, out_rate,
211 in_rate * (int64_t)phase_count, INT32_MAX / 2))
212 goto error;
213 c->ideal_dst_incr = c->dst_incr;
214
215 c->index = -phase_count * ((c->filter_length - 1) / 2);
216 c->frac = 0;
217
218 /* allocate internal buffer */
219 c->buffer = ff_audio_data_alloc(avr->resample_channels, 0,
220 avr->internal_sample_fmt,
221 "resample buffer");
222 if (!c->buffer)
223 goto error;
224
225 av_log(avr, AV_LOG_DEBUG, "resample: %s from %d Hz to %d Hz\n",
226 av_get_sample_fmt_name(avr->internal_sample_fmt),
227 avr->in_sample_rate, avr->out_sample_rate);
228
229 return c;
230
231 error:
232 ff_audio_data_free(&c->buffer);
233 av_free(c->filter_bank);
234 av_free(c);
235 return NULL;
236 }
237
238 void ff_audio_resample_free(ResampleContext **c)
239 {
240 if (!*c)
241 return;
242 ff_audio_data_free(&(*c)->buffer);
243 av_free((*c)->filter_bank);
244 av_freep(c);
245 }
246
247 int avresample_set_compensation(AVAudioResampleContext *avr, int sample_delta,
248 int compensation_distance)
249 {
250 ResampleContext *c;
251 AudioData *fifo_buf = NULL;
252 int ret = 0;
253
254 if (compensation_distance < 0)
255 return AVERROR(EINVAL);
256 if (!compensation_distance && sample_delta)
257 return AVERROR(EINVAL);
258
259 if (!avr->resample_needed) {
260 #if FF_API_RESAMPLE_CLOSE_OPEN
261 /* if resampling was not enabled previously, re-initialize the
262 AVAudioResampleContext and force resampling */
263 int fifo_samples;
264 int restore_matrix = 0;
265 double matrix[AVRESAMPLE_MAX_CHANNELS * AVRESAMPLE_MAX_CHANNELS] = { 0 };
266
267 /* buffer any remaining samples in the output FIFO before closing */
268 fifo_samples = av_audio_fifo_size(avr->out_fifo);
269 if (fifo_samples > 0) {
270 fifo_buf = ff_audio_data_alloc(avr->out_channels, fifo_samples,
271 avr->out_sample_fmt, NULL);
272 if (!fifo_buf)
273 return AVERROR(EINVAL);
274 ret = ff_audio_data_read_from_fifo(avr->out_fifo, fifo_buf,
275 fifo_samples);
276 if (ret < 0)
277 goto reinit_fail;
278 }
279 /* save the channel mixing matrix */
280 if (avr->am) {
281 ret = avresample_get_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
282 if (ret < 0)
283 goto reinit_fail;
284 restore_matrix = 1;
285 }
286
287 /* close the AVAudioResampleContext */
288 avresample_close(avr);
289
290 avr->force_resampling = 1;
291
292 /* restore the channel mixing matrix */
293 if (restore_matrix) {
294 ret = avresample_set_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
295 if (ret < 0)
296 goto reinit_fail;
297 }
298
299 /* re-open the AVAudioResampleContext */
300 ret = avresample_open(avr);
301 if (ret < 0)
302 goto reinit_fail;
303
304 /* restore buffered samples to the output FIFO */
305 if (fifo_samples > 0) {
306 ret = ff_audio_data_add_to_fifo(avr->out_fifo, fifo_buf, 0,
307 fifo_samples);
308 if (ret < 0)
309 goto reinit_fail;
310 ff_audio_data_free(&fifo_buf);
311 }
312 #else
313 av_log(avr, AV_LOG_ERROR, "Unable to set resampling compensation\n");
314 return AVERROR(EINVAL);
315 #endif
316 }
317 c = avr->resample;
318 c->compensation_distance = compensation_distance;
319 if (compensation_distance) {
320 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr *
321 (int64_t)sample_delta / compensation_distance;
322 } else {
323 c->dst_incr = c->ideal_dst_incr;
324 }
325 return 0;
326
327 reinit_fail:
328 ff_audio_data_free(&fifo_buf);
329 return ret;
330 }
331
332 static int resample(ResampleContext *c, void *dst, const void *src,
333 int *consumed, int src_size, int dst_size, int update_ctx)
334 {
335 int dst_index;
336 int index = c->index;
337 int frac = c->frac;
338 int dst_incr_frac = c->dst_incr % c->src_incr;
339 int dst_incr = c->dst_incr / c->src_incr;
340 int compensation_distance = c->compensation_distance;
341
342 if (!dst != !src)
343 return AVERROR(EINVAL);
344
345 if (compensation_distance == 0 && c->filter_length == 1 &&
346 c->phase_shift == 0) {
347 int64_t index2 = ((int64_t)index) << 32;
348 int64_t incr = (1LL << 32) * c->dst_incr / c->src_incr;
349 dst_size = FFMIN(dst_size,
350 (src_size-1-index) * (int64_t)c->src_incr /
351 c->dst_incr);
352
353 if (dst) {
354 for(dst_index = 0; dst_index < dst_size; dst_index++) {
355 c->resample_one(c, 1, dst, dst_index, src, 0, index2 >> 32, 0);
356 index2 += incr;
357 }
358 } else {
359 dst_index = dst_size;
360 }
361 index += dst_index * dst_incr;
362 index += (frac + dst_index * (int64_t)dst_incr_frac) / c->src_incr;
363 frac = (frac + dst_index * (int64_t)dst_incr_frac) % c->src_incr;
364 } else {
365 for (dst_index = 0; dst_index < dst_size; dst_index++) {
366 int sample_index = index >> c->phase_shift;
367
368 if (sample_index + c->filter_length > src_size ||
369 -sample_index >= src_size)
370 break;
371
372 if (dst)
373 c->resample_one(c, 0, dst, dst_index, src, src_size, index, frac);
374
375 frac += dst_incr_frac;
376 index += dst_incr;
377 if (frac >= c->src_incr) {
378 frac -= c->src_incr;
379 index++;
380 }
381 if (dst_index + 1 == compensation_distance) {
382 compensation_distance = 0;
383 dst_incr_frac = c->ideal_dst_incr % c->src_incr;
384 dst_incr = c->ideal_dst_incr / c->src_incr;
385 }
386 }
387 }
388 if (consumed)
389 *consumed = FFMAX(index, 0) >> c->phase_shift;
390
391 if (update_ctx) {
392 if (index >= 0)
393 index &= c->phase_mask;
394
395 if (compensation_distance) {
396 compensation_distance -= dst_index;
397 if (compensation_distance <= 0)
398 return AVERROR_BUG;
399 }
400 c->frac = frac;
401 c->index = index;
402 c->dst_incr = dst_incr_frac + c->src_incr*dst_incr;
403 c->compensation_distance = compensation_distance;
404 }
405
406 return dst_index;
407 }
408
409 int ff_audio_resample(ResampleContext *c, AudioData *dst, AudioData *src)
410 {
411 int ch, in_samples, in_leftover, consumed = 0, out_samples = 0;
412 int ret = AVERROR(EINVAL);
413
414 in_samples = src ? src->nb_samples : 0;
415 in_leftover = c->buffer->nb_samples;
416
417 /* add input samples to the internal buffer */
418 if (src) {
419 ret = ff_audio_data_combine(c->buffer, in_leftover, src, 0, in_samples);
420 if (ret < 0)
421 return ret;
422 } else if (!in_leftover) {
423 /* no remaining samples to flush */
424 return 0;
425 } else {
426 /* TODO: pad buffer to flush completely */
427 }
428
429 /* calculate output size and reallocate output buffer if needed */
430 /* TODO: try to calculate this without the dummy resample() run */
431 if (!dst->read_only && dst->allow_realloc) {
432 out_samples = resample(c, NULL, NULL, NULL, c->buffer->nb_samples,
433 INT_MAX, 0);
434 ret = ff_audio_data_realloc(dst, out_samples);
435 if (ret < 0) {
436 av_log(c->avr, AV_LOG_ERROR, "error reallocating output\n");
437 return ret;
438 }
439 }
440
441 /* resample each channel plane */
442 for (ch = 0; ch < c->buffer->channels; ch++) {
443 out_samples = resample(c, (void *)dst->data[ch],
444 (const void *)c->buffer->data[ch], &consumed,
445 c->buffer->nb_samples, dst->allocated_samples,
446 ch + 1 == c->buffer->channels);
447 }
448 if (out_samples < 0) {
449 av_log(c->avr, AV_LOG_ERROR, "error during resampling\n");
450 return out_samples;
451 }
452
453 /* drain consumed samples from the internal buffer */
454 ff_audio_data_drain(c->buffer, consumed);
455
456 av_dlog(c->avr, "resampled %d in + %d leftover to %d out + %d leftover\n",
457 in_samples, in_leftover, out_samples, c->buffer->nb_samples);
458
459 dst->nb_samples = out_samples;
460 return 0;
461 }
462
463 int avresample_get_delay(AVAudioResampleContext *avr)
464 {
465 if (!avr->resample_needed || !avr->resample)
466 return 0;
467
468 return avr->resample->buffer->nb_samples;
469 }